Towards Sustainable Forest Monitoring: Efficient Net-Based Animal Species Identification and Intrusion Detection

Abstract

Deep learning has emerged as a powerful domain for automated wildlife monitoring by enabling hierarchical feature learning and scalable real-time analysis from visual data. However, existing wildlife monitoring approaches that rely on traditional camera traps and conventional CNN-based methods often face limitations in accurately detecting and classifying species under complex forest conditions. These methods struggle with reliable recognition of visually similar species and exhibit reduced performance when identifying rare animals. To address these drawbacks, this paper proposes a Forest Surveillance System based on EfficientNet deep learning architectures. The proposed system leverages EfficientNet compound scaling to achieve efficient feature extraction with reduced computational cost, improving the recognition of both common and rare species. Additionally, the system integrates a human intrusion detection mechanism to provide early alerts for unauthorized entry into protected forest areas, thereby enhancing wildlife conservation and forest security.